ResearchAI — AI-Powered Research Agent

An intelligent, full-stack AI Research Agent that autonomously searches academic papers, generates AI summaries using IBM Granite, and enables semantic Q&A across multiple papers using MongoDB Atlas Vector Search.

---

ScreenShot

---

📌 Problem Statement

Researchers and students waste hours manually searching, reading, and synthesizing academic papers. ResearchAI solves this by automating the entire research discovery pipeline using Agentic AI.

---

✨ Features

• 🔍 Smart Paper Search — Search 200M+ academic papers via Semantic Scholar API
• 🤖 AI Summarization — Complex papers summarized in simple language using IBM Granite
• 💬 Semantic Q&A — Ask questions across multiple papers using MongoDB Atlas Vector Search
• 📌 Save Papers — Bookmark and persist important papers across sessions
• ⚡ Real-time Updates — Live agent status streaming via Socket.io
• 🗄️ Redis Caching — 60% faster repeat searches with intelligent caching

---

🏗️ System Architecture

User Query
    ↓
React Frontend (Vite)
    ↓
Node.js + Express Backend
    ↓
┌─────────────────────────────────────┐
│           Agent Pipeline            │
│                                     │
│  Search Agent → Semantic Scholar    │
│  Summarize Agent → IBM Granite      │
│  Embedding Agent → Vector Store     │
│  Storage Agent → MongoDB Atlas      │
└─────────────────────────────────────┘
    ↓
Real-time Response via Socket.io

---

🛠️ Tech Stack

Frontend

Technology	Purpose
React.js + Vite	UI Framework
Tailwind CSS	Styling
Socket.io Client	Real-time updates
Axios	HTTP requests
Lucide React	Icons

Backend

Technology	Purpose
Node.js + Express	Server & API
Socket.io	Real-time streaming
Redis	Response caching
Winston	Production logging
Joi	Request validation
Helmet	Security headers

AI & Database

Technology	Purpose
IBM Granite (ibm/granite-4-h-small)	Paper summarization & Q&A
IBM Watsonx.ai	LLM API & governance
MongoDB Atlas	Paper storage
MongoDB Vector Search	Semantic similarity search
Semantic Scholar API	Academic paper database

---

📁 Project Structure

ResearchAI/
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   │   ├── SearchBar.jsx
│   │   │   ├── PaperCard.jsx
│   │   │   ├── StatusBar.jsx
│   │   │   ├── AskBar.jsx
│   │   │   └── Sidebar.jsx
│   │   ├── pages/
│   │   │   └── Home.jsx
│   │   ├── services/
│   │   │   └── api.js
│   │   ├── App.jsx
│   │   └── main.jsx
│   └── package.json
│
└── backend/
    ├── src/
    │   ├── config/
    │   │   ├── db.js
    │   │   ├── redis.js
    │   │   └── env.js
    │   ├── controllers/
    │   │   ├── search.controller.js
    │   │   ├── ask.controller.js
    │   │   └── saved.controller.js
    │   ├── services/
    │   │   ├── arxiv.service.js
    │   │   ├── granite.service.js
    │   │   └── vector.service.js
    │   ├── models/
    │   │   ├── Paper.model.js
    │   │   └── SavedPaper.model.js
    │   ├── middleware/
    │   │   ├── error.middleware.js
    │   │   ├── rateLimit.middleware.js
    │   │   └── validate.middleware.js
    │   ├── routes/
    │   │   ├── search.routes.js
    │   │   ├── ask.routes.js
    │   │   └── saved.routes.js
    │   ├── socket/
    │   │   └── socket.handler.js
    │   └── utils/
    │       ├── logger.js
    │       ├── asyncHandler.js
    │       ├── ApiResponse.js
    │       └── ApiError.js
    ├── server.js
    └── package.json

---

⚙️ Setup & Installation

Prerequisites

Node.js >= 18
MongoDB Atlas account (free)
IBM Cloud account (free)
Redis (optional — app works without it)

1. Clone the repository

git clone https://github.com/yourusername/research-agent.git
cd research-agent

2. Backend Setup

cd backend
npm install

Create .env file:

NODE_ENV=development
PORT=5000
MONGODB_URI=your_mongodb_atlas_uri
REDIS_URL=redis://localhost:6379
IBM_API_KEY=your_ibm_api_key
IBM_PROJECT_ID=your_watsonx_project_id
IBM_GRANITE_URL=https://us-south.ml.cloud.ibm.com
JWT_SECRET=your_jwt_secret
CLIENT_URL=http://localhost:5173

npm run dev

3. Frontend Setup

cd frontend
npm install

Create .env file:

VITE_API_URL=http://localhost:5000

npm run dev

4. Open Browser

http://localhost:5173

---

🔌 API Endpoints

Method	Endpoint	Description
POST	/api/search	Search papers by topic
POST	/api/ask	Ask question across papers
GET	/api/saved	Get all saved papers
POST	/api/saved	Save a paper
DELETE	/api/saved/:arxivId	Remove saved paper
GET	/health	Health check

---

🌐 Environment Variables

Variable	Description
MONGODB_URI	MongoDB Atlas connection string
IBM_API_KEY	IBM Cloud API key
IBM_PROJECT_ID	Watsonx.ai project ID
IBM_GRANITE_URL	IBM Granite endpoint URL
REDIS_URL	Redis connection URL
JWT_SECRET	JWT signing secret
CLIENT_URL	Frontend URL for CORS

---

🔒 Security Features

• Helmet.js — HTTP security headers
• Rate limiting — 10 searches/min per IP
• MongoDB sanitization — prevents injection attacks
• Input validation — Joi schema validation
• Environment variables — no secrets in code
• Error handling — stack traces hidden in production

---

🎯 IBM SkillsBuild Internship Project

This project was built as part of the IBM SkillsBuild University Engagements — AICTE 2026 internship program.

Problem Statement: No. 1 — Research Agent

IBM Technologies Used:

• IBM Watsonx.ai — Model deployment and API access
• IBM Granite (ibm/granite-4-h-small) — LLM for summarization
• IBM Cloud Lite — Free tier infrastructure

---

👩‍💻 Developer

Palak Verma B.Tech CSE — CSJM University, Kanpur (2024-2028)

---